I added a try/catch block so that any error that arises within a
cmd() call is also reported on the log file (if open).

In principle, this should allow removing all the duplicate code where we
both write on the log and on the plumed_merror.

Notice that in order to satisfy astyle the whole content of PlumedMain::cmd()
has been indented so appears as changed in this commit. Actually, it is only
indented.

This commit closes #319

I removed the word "Internal" since it looks like we are using this path
to report also errors in the input files.

Now exceptions are always enabled. As a consequence, I removed the requirement
for "exceptions" in one regtest. Notice that this means that plumed v2.5
would skip this regtest as it is in v2.4.

Using in-class constructor

Harmless error introduced in 3d93bfd2

I noticed that cache is not saved if brew is removed
(apparently some travis tool requires a brew library)

ERMSD now gives the same results irrespectively of units.

The value of PLUMED_ROOT stored in statically patched MD code was
incorrectly set to the compilation directory rather than the installed one.

The only way to trigger the bug from regtest was by observing the behavior
of rt-make-exceptions when plumed was installed by macports (since in this
case the compilation directory is removed).

Messages related to NOPBC keyword were wrong

Thanks to @valsson

- added some info in the error message of the parser
- std::string everywhere

Includes mention to ABIN code

warning on periodicity for grids is now an error becasue the grid spacing can be completely wrong

- added default value for r_0
- simplified examples